Area and power-efficient hardware implementations are crucial for the widespread adoption of post-quantum cryptography (PQC) algorithms like FALCON. One of the main operations in FALCON is the Number Theoretic Transform (NTT), which needs to be performed with many prime numbers. Having one set of twiddle factors (TFs) for each prime makes storing all of these TFs impractical. In this paper, we propose an architecture for generating TFs on the fly for FALCON-oriented NTT, designed for area and power efficiency. Our approach dynamically generates TFs during NTT computations, significantly reducing on-chip memory requirements. The ASIC implementation results demonstrate significant improvements, with the proposed design reducing on-chip memory requirements by 99%, occupying 95% less area, and consuming 87.4% less power compared to the traditional ROM-based implementation. Furthermore, our design achieved a much higher maximum clock frequency, indicating superior performance in accessing twiddle factors. These findings highlight the potential of our proposed architecture for efficient hardware implementations of FALCON-based cryptographic systems.

The global shift towards renewable energy (RE) sources, driven by climate change and fossil fuel depletion, encounters challenges in integrating intermittent sources like wind and solar power into existing grids. Developing countries, such as Nigeria, with weak grids, face significant limitations on RE penetration. This study focuses on enhancing RE grid integration in Nigeria by strategically deploying Flexible Alternating Current Transmission System (FACTS) devices. Results obtained from the Nigerian grid indicate that FACTS devices, particularly STATCOMs, can extend the penetration limit by 40%, enabling an additional 152 MW of wind energy to be integrated without jeopardizing system stability. Improved voltage profiles and enhanced stability highlight the effectiveness of FACTS devices in facilitating RE integration into weak grids. Thus, developing grids can accommodate more renewable energy without extensive reconfiguration of the power system architecture.

Doppler shift has significant implications in Low Earth Orbit (LEO) constellations due to the rapid relative movement of satellites. This paper presents a novel approach based on stochastic geometry to capture the statistical distribution of Doppler shift in modern mega satellite constellations. We derive a simplified expression for both the Doppler shift s-curve and its maximum value, highlighting the dependency of Doppler shifts on the zenith angle of the closest approach during a satellite pass. Additionally, we present a stochastic model for the distribution of these zenith angles within the random constellation model. Furthermore, the statistical distribution of the Doppler bound is analytically formulated and validated through extensive Monte Carlo simulations of two major satellite constellations currently in orbit.

In the present study, we focus on the object handover task as a major example of collaborative work between a human and a robot. To achieve a smooth handover between two different agents, their mutual communication is indispensable for understanding the other’s intentions. However, previous research has not dealt with a moment during handover in which a robot takes the object from a human grasp or in which a robot hands the object to the human. It should be noted that the performance during those phases is crucial to the success or failure of the tasks because slight changes in the kinematics of the relationship between hands or fingers and object result in significant changes in grasping status: the human may forcibly pull out the object while the robot is grasping it or drop the object. Therefore, this study aims to realize a smooth handover between a human and a robot, focusing on the moment of object handover. To this end, this paper proposes to present the stiffness of the robot hand to the human. We conducted the subject experiments to investigate the effect of this method on humans in the human-robot interaction. Experimental results show that this presentation method enables the worker to recognize the stiffness of the robot, which is difficult to recognize visually, thereby reducing the workload and allowing the worker to respond seamlessly to changes in the robot’s stiffness.

Pulsatile physiological signals, characterized by rhythmic fluctuations, are vital for assessing health conditions and are widely used in wellness devices and medical equipment. Despite their significance, models addressing domain-specific unmet needs and considerations have not been developed as much as in other fields. Therefore, building on the foundation of variational autoencoders, we introduce VABAM, a novel model for the amplitude-based synthesis of pulsatile physiological signals. The uniqueness of VABAM lies in its ability to maintain the morphological identity of signals throughout the synthesis process, achieved by integrating pass filter effects within the variational autoencoder architecture. To assess the effectiveness of the model, we developed three novel metrics based on joint mutual information. These metrics were aimed at evaluating the disentanglement of latent spaces, influence of ancillary information on signal morphologies, and controllability of amplitude-based synthesis within morphological identities. Comparative analyses demonstrated that VABAM and its variants were notably effective at preserving morphological integrity, highlighting their potential to minimize morphological distortions in physiological signal processing and their compatibility with artificial intelligence models employing frequency and amplitude features. Additionally, the proposed metrics, compatible with probabilistic models, were empirically proven to capture the characteristics of various models from multiple perspectives.

Under the commonalities found in the goals of two areas, neural network compression and feature selection for dimension reduction, this research focused on finding a new method to address both issues: a method that can lead to easier feature selection, and an enhancement in the capacity of information flow control of neural network compression techniques, especially clustering based compression. Specifically, this research focused on creating a novel and effective framework to transform the weight matrix between the input layer and the first hidden layer in neural networks to be optimal. In other words, a method that can make the weight matrix's structure itself optimal for information extraction. By proposing a simple, yet powerful weight clipping + GMM based method called an In-and-Out Weight Box that can intrinsically act similar to filtering while increasing the possibility of getting better results in compression, the main aim of research was found to be satisfied. Using Glioma Grading data from the UCI Repository for checking performance of the In-and-Out Weight box in fitting neural networks, it was found that significantly better compression results can be achieved in terms of weight sharing via clustering. This research also suggests a new feature selection method based on the In-and-Out Weight box constraint called IOW-FI, which can lead to solving limitations or problems of filtering techniques such as setting the number of components to be selected as efficient features or considering joint distributions of feature space.

Simplicity is held by many to be the key to general intelligence. Simpler models tend to “generalise”, identifying the cause or generator of data with greater sample efficiency. The implications of the correlation between simplicity and generalisation extend far beyond computer science, addressing questions of physics and even biology. Yet simplicity is a property of form, while generalisation is of function. In interactive settings, any correlation between the two depends on interpretation. In theory there could be no correlation and yet in practice, there is. Previous theoretical work showed generalisation to be a consequence of “weak” constraints implied by function, not form. Experiments demonstrated choosing weak constraints over simple forms yielded a 110-500% improvement in generalisation rate. Here we show that all constraints can take equally simple forms, regardless of weakness. However if forms are spatially extended, then function is represented using a finite subset of forms. If function is represented using a finite subset of forms, then we can force a correlation between simplicity and generalisation by making weak constraints take simple forms. If function is determined by a goal directed process that favours versatility (e.g. natural selection), then efficiency demands weak constraints take simple forms. Complexity has no causal influence on generalisation, but appears to due to confounding.In Press: Accepted for publication in the Proceedings of The 17th Conference on Artificial General Intelligence, 2024

This paper undertakes a critical examination of unsupervised learning within the context of Raven's Progressive Matrices (RPMs). We trace the historical trajectory of computational models for RPMs, from early rule-based approaches to modern neural networks, and we focus on the innovative work of Zhuo et al. in introducing semi-supervised learning to RPMs. Our discussion highlights the nuances of unsupervised learning, emphasising the role of noisy labels as a form of guidance, albeit with a trade-off in precision compared to traditional supervised learning. In this paper, we recognise the challenge in formalising the distinction between supervised and unsupervised learning, but we underscore the importance of precision in communication and nomenclature, especially in regards to facilitating knowledge transfer and directing future research. We hope that this contribution enhances the discourse on unsupervised learning and offers valuable insights towards the challenges and opportunities in attaining human-level reasoning capabilities in machine learning and artificial intelligence.

In the contemporary digital landscape, online learning has emerged as a widely embraced mode of education due to its accessibility and cost-effectiveness. Timely delivery of courses tailored to learners’ needs is pivotal in maintaining their focus and commitment to learning. This project highlights the significant role of recommendation systems in attracting learners to focus on their learning paths and achieve their goals. It underscores the economic accessibility and the opportunities for individuals to engage with essential career skills. These concerns are particularly salient given the high cost of traditional education and the perceived elitism of certain institutions, which create barriers for economically disadvantaged learners.The project employs a multi-faceted approach to enhance recommendation accuracy. Initially, it collects user information to gauge Feature Ratings, reflecting learners’ reactions to courses. Additionally, course descriptions undergo analysis to ascertain word importance and inter-course relevance. While a standalone content-based approach proves insufficient, the project adopts a collaborative filtering approach next. Here, Feature Ratings are learned through a K-Nearest Neighbors (KNN) model, leveraging a similarity metric to identify similar learners. Crucially, the project integrates these methodologies into a Hybrid Filtering model, combining the strengths of both approaches for optimal performance. Performance evaluation showcases the system’s efficacy, measured through Hit Rate and F1 Score accuracies, demonstrating its effectiveness in enhancing learning outcomes.

Machine-Learning model implementation in Resource-Scarce Embedded Systems is becoming a standard in many systems and projects. This implementation allows systems to be less Cloud dependent and make decisions independently. As these systems' reasoning becomes intricate with the Machine Learning model's decision, an attack to change the Machine Learning model's data structure can make the entire system misbehave, which in some solutions can be critical. Therefore, it is necessary to create low-overhead tools to flag any miscalculation or wrongdoing during the model's inference phase. The following work presents a Machine Learning Health Monitoring system based on PCA and Control Charts to verify if the model's inference function runs properly. The solution presents a reasonable flag rate and is implemented e ciently in a Resource-Scarce Embedded System due to its reduced memory and processing footprints.

Global Navigation Satellite System (GNSS) data has become a stable of modern life, whether delivered through a smartphone, vehicle, or as part of an industrial dataset. The expectation of reliable position and time information, irrespective of the environment, becomes challenging in tunnels and canyons with significant multipath, whether natural or urban. An open dataset of Lidar, Computer Vision, Inertial measurement, and GNSS data is presented, combined with preliminary analysis of the dataset on a selected recording, considering GPS coarse acquisition, lidar and computer vision based navigation together with model based channel reconstruction. Lidar and video based techniques demonstrate positioning error as low as 3.8m root mean square error, together with GNSS acquisition and signal blockage alignment with computer vision and lidar maps.

Android is one of the most popular mobile operating system (OS). Android is currently developed by the Google and other developers. It is an open source OS, based on the Linux kernel. Android can be used on smartphones, tablets, smartwatches and other devices. Android supports different types of applications, hardwares and firmwares. The Android operating system has different flavors and versions due to its open nature. This can lead to security and privacy concerns. Google Play Store is the most popular online app store for Android. As of 2015, more than 1.6 million applications have been listed on the Google Play Store [1]. In this project, I develop 'SecuFone: Android Security Advisor' an Android application. SecuFone evaluates the security of the overall system and provides advisory (recommendation) to improve their security. It evaluates security by using a hierarchical system. Hierarchical system uses fuzzy logic to provide users with the generalized score based on the metrics. The recommendations and score generated by the application helps users to enhance their privacy and security of their devices.

Fourier transform an algorithm which is used to decompose pure analogue signals has very few practical applications which led to the discovery of Discrete Fourier transform (DFT), even though DFT considered the finite, discontinuous nature of real-world signals, its application was limited due to the requirement of high calculating power to decompose the signal. Discovery of Fast Fourier transform (FFT) in 1965 reduced the number of calculations and calculation time required to decompose a signal and revolutionised the world with numerous applications such as radar, sonar, 5G, Wi-Fi, seismology making it a cornerstone in signal processing. We discuss the historic significance of discovery of FFT, its significance in nuclear arms race, overcoming the challenge of detecting covert underground nuclear tests, FFT algorithm helping in adoption of Comprehensive Nuclear Test Ban Treaty (CTBT) and the missed opportunity to stop nuclear weapon development by earlier discovery of the same algorithm.

The inconsistency between training strategies and task objectives is a principal constraint impeding DensePose development. DensePose aims to minimize the surface distance between the estimations and the ground truth, thereby establishing a correspondence between two-dimensional images and threedimensional humans. However, the discrete IUV optimization strategy employed by the existing pipeline deviates from the goal of minimizing surface distance, thus yielding sub-optimal results. To solve this problem, we propose Geodesic-consistent RCNN (GC RCNN), which models the intrinsic interdependence within the IUV to enhance human surface understanding and facilitate surface distance optimization. GC RCNN incorporates the adaptively semantic enhancement module (ASEM) and the geodesic-consistent loss (GCL) into the traditional DensePose estimation framework. ASME enhances human surface semantics by extracting distinct IUV features and dynamically modeling the feature-level IUV interdependence. GCL extracts the surface distance information by quantifying the degree of deviation between the estimated IUV distribution and the ground truth. The distribution deviation is utilized to refine the IUV optimization process, thereby aiding in minimizing surface distance. Empirical analyses conducted on the DensePose-COCO dataset validate the superior performance of GC RCNN, which surpasses the base model by a margin of up to 3.3% AP.

This paper designs novel nonparametric Bellman mappings in reproducing kernel Hilbert spaces (RKHSs) for reinforcement learning (RL). The proposed mappings benefit from the rich approximating properties of RKHSs, adopt no assumptions on the statistics of the data owing to their nonparametric nature, require no knowledge on transition probabilities of Markov decision processes, and may operate without any training data. Moreover, they allow for sampling on-the-fly via the design of trajectory samples, re-use past test data via experience replay, effect dimensionality reduction by random Fourier features, and enable computationally lightweight operations to fit into efficient online or time-adaptive learning. The paper offers also a variational framework to design the free parameters of the proposed Bellman mappings, and shows that appropriate choices of those parameters yield several popular Bellman-mapping designs. As an application, the proposed mappings are employed to offer a novel solution to the problem of countering outliers in adaptive filtering. More specifically, with no prior information on the statistics of the outliers and no training data, a policy-iteration algorithm is introduced to select online, per time instance, the "optimal" coefficient p in the least-mean-p-power-error method. Numerical tests on synthetic data showcase, in most of the cases, the superior performance of the proposed solution over several RL and non-RL schemes.

In this work, we utilize symmetric Fourier features to improve the accuracy of deep learning-augmented MOSFET current-voltage model. We first apply the Fourier analysis on the cost function and discuss the spectral bias of neural network. The benefits of augmenting the neural network with symmetric Fourier features, designed to satisfy the Gummel symmetry, are then discussed. Using TCAD-simulated FinFET as our dataset, our improved model, augmented with 8 Fourier features, achieves outstanding 3-sigma errors of 0.15% for IV characteristics, 4.14% for output conductance, and 0.33% for transconductance. This represents a cost reduction of 13 times compared to the scenario without the Fourier features, highlighting the benefits gained from their inclusion.

Text generation is an important method to generate accurate and available product description from product title. Product description generation's main problem for online E-commerce application is the available rate of generated text. The available rate of online deployment standard needs to reach above 99%. Model-centric method is limited by the quality of the training dataset. To handle the problem, we propose our data-centric method to improve the generation model's available rate from 88.0% to 99.2%. Our approach helps in building models using LLMs (large language models) annotation results and constructing datasets to obtain better results than LLMs. Also, our method simplifies the human labeling work to 2-class choices to label, which improve the labeling speed. In summary, our method saves about 10x of the labeling time and achieves 99.2% accuracy to be deployed online.

The phenomenon of hallucinations takes place when generative artificial intelligence systems, such as large language models (LLMs) like ChatGPT, generate outputs that are illogical, factually incorrect, or otherwise unreal. In generative artificial intelligence, hallucinations have the ability to unlock creative potential, but they also create challenges for producing accurate and trustworthy AI outputs. Both concerns will be covered in this abstract. Artificial intelligence hallucinations can be caused by a variety of factors. There is a possibility that the model will show an inaccurate response to novel situations or edge cases if the training data is insufficient, incomplete, or biassed. It is common for generative artificial intelligence to generate content in response to cues, regardless of the model's "understanding" or the quality of its output. The rising body of material will be synthesised in the paper in order to provide a comprehensive approach to AI hallucinations. The purpose of this study is to gain a deeper understanding of artificial intelligence hallucinations and to develop strategies that can reduce the negative impacts of these hallucinations while simultaneously maximising their creative potential. For the purpose of contextualising artificial intelligence hallucinations in generative AI and providing information for the building of more reliable AI systems, the technique will examine models and mind maps from the journals that were referred to. Hallucinations experienced by AI models can be mitigated through the critical analysis of AI outputs and the diversification of data sources. This paper stresses the significance of high-quality training data, human feedback, transparency, and ongoing quality control in the development of artificial intelligence. This is accomplished through a synthesis of the relevant literature as well as an analysis of models and mind maps.